(ii) Parasite sharing (Λ)

We considered three different methods of calculating parasite sharing to examine the robustness of results to variation in sharing estimates (figure 1b). A secondary goal of this analysis was to also identify methods that may be more practical in resource-limited contexts. First, we used a Bray-Curtis dissimilarity matrix calculated from a DNA metabarcoding network of strongylid parasites in herbivore dung collected from Mpala Research Center, less than 40 km away [10]. Specifically, the distance matrix was calculated using the pairwise dissimilarity of mean parasite read abundances per host species. Because metabarcoding cannot differentiate between infective parasites and low levels of parasites that are consumed and passed by hosts without infecting them, we excluded mean values less than 2%, as described in [10]. We also excluded one case of an exclusively equid parasite, Cylicostephanus minutus, that was found at 2.1% relative read abundance in giraffes. We then converted dissimilarities to a similarity matrix by subtracting the scaled distance matrix from 1. Second, we retrieved host-parasite records from the London Natural History Museum's database [34] for the focal species in our study using the helminthR package [35]. We restricted the search to faecal-orally transmitted nematode parasites identified to species. We then computed Jaccard similarities for each pairwise combination using the vegan package [37]. Third, we pruned a mammal phylogenetic tree [36] to species in our study and computed all pairwise phylogenetic distances. We then scaled this distance matrix from 0 to 1, with the maximum distance (elephant – giraffe) set to 1. Finally, to give more closely related species a higher probability of parasite sharing than more distantly related species, we subtracted the scaled distance matrix from 1.